Hybrid Long-Distance Functional Dependency Parsing

We propose a robust, hybrid, deep-syntactic§dependency-based §parser and present its implementation and evaluation. §The parser is designed to keep search-spaces small§without §compromising much on the linguistic performance or§adequacy. §The §resulting parser is deep-syntactic like a formal§grammar-based §parser while mostly context-free and fast enough for§large-scale §application. It combines successful current§approaches into a §hybrid, modular and open model. We suggest,§implement, and §evaluate a parsing architecture that is fast, robust§and efficient §enough to allow users to do broad-coverage parsing of§unrestricted §texts from varied domains. We present a probability§model and a §combination between a rule-based competence grammar§and a §statistical lexicalized performance disambiguation§model. We treat §long-distance dependencies with post-processing and mild §context-sensitivity. We conclude that labelled§Dependency Grammar §is sufficiently expressive for linguistically§adequate parsing. We §argue that our parser covers the middle ground§between statistical §parsing and formal grammar-based parsing. The parser has §competitive performance and has been applied widely.